Augmented Reality (AR) devices are being increasingly used for just-in-time informatics that can assist humans in completing a variety of tasks in Future of Work scenarios. An effective AR system can help front-line workers both to adapt their skills to new domains and to work safely in new environments. While AR devices have matured in their software and hardware capabilities, the problem of determining and delivering the right AR content in a timely manner to ensure accuracy and improve the efficiency of human operators remains challenging. This paper describes the application of the DDDAS paradigm in the design of a new suite of edge/cloud-hosted, AR content-driven AI/ML tools and services that can support humans to cope with unanticipated and unexpected situations. The paper then validates our solution on a newly created dataset for car maintenance and discusses the adversarial robustness issues that arise in AI/ML approaches.